Monarch
From connected tools to governed outcomes.
Working with Monarch feels like briefing a capable team: you connect the tools, share the context, describe the outcome, and keep sign-off on the moments that matter. Here's the loop — and what it looks like on real work.
Six steps, repeated
Connect your tools
Link email, CRM, calendars, storage, and thousands of other app actions through managed connectors — shared across the organization or owned by individuals. Each connection exposes specific, named actions, not blanket access.
Set the rules of engagement
For every action on every connection, choose an approval mode — run automatically or wait for a human — plus risk levels and role restrictions on who may approve. Governance is configured once and enforced everywhere.
Ground it in your knowledge
Upload documents and organize them into search groups with tuned retrieval — chunking, embedding models, reranking — so agents answer from your material with the retrieval quality each collection deserves.
Ask — or let events ask for you
Start work from chat in plain language, or let inbound webhook events start it for you. Either way a durable job is created, planned by Master, and fanned out to Task specialists.
Approve what matters
Watch plans and progress stream into the thread. When a consequential action is proposed, a sign-off card appears in your Activity queue with full context — approve it, deny it, or step in and adjust.
Review and improve
Outcomes, costs, and complete histories are queryable after every run. Approve the same action enough times and you'll know it's safe to automate — flip the mode and the pipeline gets faster.
What teams run on Monarch
Patterns proven in production-shaped deployments — each one a supervised pipeline with humans on the consequential moments.
Lead qualification & follow-up
Sync a lead bank from CSV and CRM sources with change detection and conflict tracking. Agents enrich each record, score it with a written rationale, and recommend next actions. Inbound signals — like a sample request — trigger follow-up jobs automatically, and outreach drafts wait for human approval before a single email leaves the building.
Grounded research & answers
Turn scattered documents into a governed knowledge repository. Teams ask questions in chat and get answers grounded in their own material — with retrieval tuned per collection, embedding status visible per document, and access scoped to the organization and the approved users within it.
Pipelines that start themselves
Provider webhooks arrive as signed, idempotent events, get deterministically correlated to the records they concern, and launch jobs — no polling, no duplicates. The same governance applies whether a person or an event started the work.
Autonomy is a dial, not a switch
Monarch doesn't ask you to choose between "fully manual" and "fully autonomous." Approval modes are set per action, per connection — so a pipeline can research freely, write to internal systems automatically, and still stop for sign-off before customer-facing sends.
- Start conservative. New connections default to human review on consequential actions.
- Earn autonomy. When an action has proven itself, flip it to automatic — the policy result is still recorded on every run.
- Scope by role. Restrict who may approve or use sensitive actions to the right organization roles.
- Change instantly. Modes are data, not code — adjust governance without redeploying anything.
Automatic
Routine, low-risk actions clear policy and run unattended — recorded like everything else.
Human approval
Consequential actions wait for sign-off with full context in the Activity queue.
Risk levels
Each action carries an assessed risk level that informs policy and reviewers.
Role gates
Approval rights follow your org chart — admins decide who decides.
Brief your first pipeline.
Connect a tool, set the rules, and hand Monarch something real.